Performance Evaluation of the EXT4 File System A Comparative Study Against EXT3, ReiserFS and JFS Alireza Ghobadi1, Amir Hesam Yaribakht1, Sanam Maham2, Mohammad Hossein Ghods2 Faculty of Information Technology Multimedia University Cyberjaya 63000 Selangor, Malaysia

1

2

SOHA Sdn. Bhd.,AB1 MSC Center, Cyberjaya, Selangor, Malaysia

Abstract— Base on several definition of File System, File System is a system which is responsible to handle files and managing data in any operating system [1]. According to these definitions, choose a File System to managing data on your system, is one of the consideration for anybody who use a computer. Anybody can fill this problem when you have a lot of files with a large portion. File System types can be classified into disk File Systems, network File Systems and special purpose File Systems[2][3][4.] The purpose of this Paper is “comparison Performance evaluation among four File Systems”. According to above goals, there are several file systems on Linux operation system. These file system are EXT2, EXT3, EXT4, ReiserFS, JFS, and etc. In this paper, we are neither define several file systems on Linux nor compare them and tested by IOzone and Postmark benchmark tools [5][6]. The performance result has shown base on read, reread, write and re-write of file for IOzone benchmark tool and create, read, append, delete for Postmark benchmark tool. We have chosen IOzone bench mark for our experiments as it is preferred for operating system evaluation [7]. Keywords- Performance Evaluation, File System, EXT4, EXT3, ReiserFS

According to above goals, there is several file system on Linux operating system. These file system are EXT2, EXT3, EXT4, ReiserFS, JFS, and etc [7][8][9][10][11][12]. In this paper, we are neither define several operating systems on Linux nor compare them and tested by IOzone and Postmark benchmark tools. We will be showing performance of them based on read, re-read, write and rewrite of file. We have chosen IOzone and Postmark benchmark tools for our experiments as it is preferred for operating system evaluation [13][14].

1.1

This paper, compares and evaluates EXT4 against EXT3 and ReiserFS, JFS file systems on Linux. The paper studies on these four file system in section 2, 3, 4. Related work in section 5. Then, it focuses on performance evaluation tools in section 6. Implementation of benchmark tools in section 7. Results and discussion in section 8, and finally, conclusion and future work in section 9.

2 1

Introduction

Base on several definition of file system, File system is a system which is responsible to handle files and managing data in any operating system [1]. According to these definitions, choose a good file system with the best performance is one of the considerations [1]. File system types can be classified into disk, network and special purpose. A disk file system designed for the file storage on a data storage device. Normally disk drive, directly or indirectly connected to the computer or any computer device. The principal aims of file system are to address scalability, performance, reliability, capabilities and robustness. On the other hand, the most popular Linux file system due to its reliability, rich feature set, relatively good performance, and strong compatibility between versions of file system.

Instructions for authors

Extent file systems

Extent file system was earlier file system that developed on Linux by Rmy Card, Laboratoire. Until now Extended has four generations and implement on several version of Linux [10]. EXT2 designed by Wayne Davidson and Stephen Tweedie and Theodore. They extended from EXT file system which designed byRemy Card and implement on the standard Linux file system [15][16]. EXT3 is one of the traditional Unix-derived file systems. it used a indirect block mapping scheme to keep track of each block. It the same data structures and supports journaling. EXT3, which just added some features to EXT2 while keeping the on-line format and approach of EXT2 [9]. EXT4 was developed by Theodore Ts’o, who was, at the time In 2006. The uber Linux developer, which developed the EXT3 maintainer, began work on EXT4., EXT4 changed a deep code change and the data structures. These changes used

to make a better file system, faster, reliable, more features, and better code. The most important and hard working on EXT4, added new features such as, Extents, journaling check summing, block allocation, delayed allocation, faster fsck, online defragmentation, and larger directory sizes (up to 64,000 files) [12].

3

ReiserFS

ReiserFS was developed as a part of the standard Linux kernel by Hans Reiser. [2] It is available on the most version of Linux operating system. ReiserFS supports metadata journaling. The ReiserFS has excellent performance for smallfiles. ReiserFS Developed based on B* Balanced Trees to organize directories, files, and data. B* provides fast directory lookups and fast deletes operations. Other performance features include support for sparse files and dynamic disk inode allocation [2].

4

Journalin File Systems (JFS)

JFS introduced by IBM as UNIX file system with the initial release of AIX Version 3.1. It has now introduced a second file system that is to run on AIX systems called Enhanced Journal File System (JFS2). JFS2 is available in AIX Version 5.0. The JFS Open Source code on originated [17]. JFS is modified primarily for the high throughput and reliability requirements of servers. JFS uses extent-based addressing structures, along with clustered block allocation policies. It is make compact, efficient, and scalable structures for mapping logical offsets within files to physical addresses on disk. An extent is a sequence of contiguous blocks allocated to a file as a unit. The addressing structure is a B+Tree populated with extent descriptors, rooted in I-node and keyed by logical offset within the file [17]. JFS supports block sizes of 512, 1024, 2048, and 4096 bytes on a per-file system basis. Smaller block sizes reduce the amount of internal fragmentation. However, small blocks can increase path length since block allocation activities may occur more often than if a large block size was used. The default block size is 4096 bytes [18]. JFS supports both sparse (which allow data to be written to random locations within a file without instantiating others unwritten file blocks.) and dense files, on a per-file system basis. [18]

5

Related Work

Some researcher studied on Extents file systems.[12][13][14] Avantika Mathur have worked on EXT3. The purpose of their research was to provide branch of EXT4 from EXT3.[4] They compared EXT3, EXT4; XFS file systems with three tools as FFSB, IOzone and Postmark. With FFSB tool they test these file systems base on throughput (MB/Sec) and CPU percent usage. [13] [14] This test has shown that XFS has higher throughput (MB/Sec) than EXT3 and EXT4, but it has lower CPU percent

usage than EXT3 and EXT4. EXT4 has higher throughput (MB/Sec) performance than EXT3, but it has lower CPU percent usage than EXT3 [9]. By IOzone tool they test these file systems base of six operations as write, re-write, read, reread, random write and random read [11]. In this test shows that in Write, re-write, random write and random read, EXT4 has higher throughput (KB/Sec) performance than XFS and in general EXT4 has higher throughput (KB/Sec) performance in all six operations than EXT3[9]. The test also shows that in read and re-read operations XFS has higher throughput (KB/Sec) performance than EXT3 and EXT4. Also observe that in re-write, random write, random read, XFS has higher throughput (KB/Sec) performance than EXT3. In write operation EXT3 has higher throughput (KB/Sec) performance than XFS. With Postmark tool they test these file systems based on two operations as read and write. The test result not only shows that EXT4 has higher throughput (MB/Sec) performance than EXT3 and XFS, but also EXT3 has higher throughput (MB/Sec) performance than XFS. In their comparison they find that EXT4 has a good improvement of EXT3 and has become an enterprise-ready solution, with a good balance of scalability, reliability, performance and stability. [2] Other researcher named, Ricardo Galli works on journal file systems available for Linux as EXT3, ReiserFS, XFS and JFS and they introduce to the basic concepts of file systems, buffer-cache, and page-cache carried out in the Linux kernel. [17]Their performance result shows that XFS, ReiserFS and EXT3 have demonstrated that they are excellent and reliable file systems. In this research, they achieved (i) EXT3 is going to be the standard file system for Linux operating system, specially Red Hat, (ii) JFS is a valid alternative for migrating AIX and OS/2 installation to Linux. (iii) In all journal file systems, ReiserFS is the only file system which has standard Linux tree since 2.4.1 which SuSE supports it. (iv) XFS is being used in large servers (especially in the Hollywood industry). It is due mainly to the influence of SGI market. (v) JFS has gotten the worst results (when tested by any benchmarks) not only on performance, but also for stability issue in the Linux port. Dr. Oliver Diedrich has a well done study on EXT3 and EXT4 file systems. He compares the structure of EXT3 and EXT4 file systems base on large volumes, huge files and extent trees. He evaluates a performance of these two file systems based on creation (based on time and write speed) and deletion (based on time) of eight 1 GB files and 10000 random read and write operations in 8 GB. He did not mention what tools he used in his test, the performance with large files between EXT3 and EXT4. The tests shows that in creation of eight 1 GB files, time in EXT4 improved 6.9% and write speed also improved 7.0% than EXT3. In deletion of eight 1 GB files, time improved 97.2% than EXT3. And among 10000 random read and write operations in 8 GB, EXT4 improved 10.9% than EXT3[11].

6

Performance Evaluation Tools

 Now searching is in process and gives the latest model of the searched benchmark.

Benchmark is a tool for performance evaluation. There are several benchmarks for file system available [13] IOzone is one of the famous benchmark tools on file system to generate and measures a selection of file operations. It has been runs for test many operating systems. The IOzone tests file I/O performance. I/O performance tests based on Read, re-read, read backwards, read strided, write, re-write, fread, fwrite, random read/write, pread/pwrite variants, aio_read, aio_write, mmap.It is useful for file system analysis of a vendor’s computer platform. [13] Another famous benchmark is The Postmark. It is responsible to creating a massive bulk of alternatively modifying files and calculating the transaction rates for a workload approximating a large Internet electronic mail server [14]. Postmark operation, produces random text files. The text files size categorizes from low bound to high bound. The size is configurable between low and high bound The text file pool is of configurable size and can be located on any accessible file system.[14] Once the bulk has been created consists of a pair of smaller transactions (i) Create file or Delete file and (ii) Read file or Append file operation. According to comparison of these benchmark tools, the achievement the IOzone performance shows that, this tool is more suitable for experimental result due to the performance of IOzone is higher than Postmark tool.

7

Implementation

File system benchmarking requires careful setup. An issue one must often contend with is how to defeat the effects of the file system buffer cache. Without careful experimental design, all of the file system requests could be satisfied in the cache and no disk activity would occur. A usual way to avoid this problem is to use a total file size that exceeds the amount of main memory available on the system. Another approach is to use a file-access mode that bypasses the file system buffer cache, such as O_DIRECT. We chose to not use O_DIRECT for this paper The second issue that one must address is estimating the accuracy of the results of the test. In our experience, file system benchmarks are notorious for being non repeatable, bimodal, and full of hysteresis effects, making it a challenge to get consistent results. In this paper we have IOzone and Postmark benchmarks using below system:  The small system that is Intel(R) processor Core(TM) Duo 2.20 GHz with 4GB of memory and a 320 GB SATA disk. For the experiments of this paper, this machine was booted with 4 GB of RAM.  Nowadays, mentioned machine provides a sampling used to run Linux.  For using IOzone and Postmark benchmarks firstly we should install them in the system.

8

Results and Discussion

This section is a description of test result. Each graph defines system’s structure. 8.1

IOzone result

Read graph shows that EXT4 has not good performance in those file size less that 100 MB. EXT4 is Extent base allocation. It is block allocation and it use contiguous allocation to allocate the file in blocks of disk. Because of this type of allocation, EXT4 has overhead and performance is not good on small file. In other hand, small file size because it is contiguous when file finish there are some space in the contiguous file that still empty. The other reason is the overhead for read the file system should refer to directory that has file name. Address of file and length of the file and if the file be small, this cause overhead happened. [13] Allocation features use for large file size. Figure 4 shows that the performance of EXT4 is higher than other file systems. For the large file, EXT4’s performance is higher than other file systems because of block allocation that fill the contiguous block in disk. EXT3 allocates blocks for a file one at a time (typically using 4KB blocks). For very large files, the associated function that doses the allocation will have to be called thousands of times. EXT4 uses ―multi-block allocation‖. It allows multiple blocks (hence the name) to be allocated during one function call. This can greatly improve the performance of EXT4 relative to EXT3, particularly for large files. [12] JFS dynamically allocate space for disk I-nodes as required, freeing the space when it is no longer needed. Two different directory organizations are provided. (i) The first organization is used for small directories and stores the directory contents within the directory’s I-node. This eliminates need for separate directory block I/O as well as need to allocate separate storage. [18] By using directory’s Inode can eliminate separate directory block I/O and allocate separate storage. (ii) Organization is used for larger directories and represents each directory as a B+Tree keyed on name. It provides faster directory lookup, insertion, and deletion capabilities. Because of these reasons JFS in small file has better performance than EXT4 but for file size from 100MB and above JFS has lower performance than EXT4. [12] ReiserFS uses B* Balanced Trees to organize directories, files, and data. This provides fast directory lookups. [2] So it has better performance on read operation in small file size generally less and equal than 20MB file size. First graph shows result base on read feature. Figure1 shows that experience by IOzone benchmark.

Figure2 shows that EXT3, ReiserFS, and JFS has a good performance when file size is small. But with increasing file size, JFS and EXT3 performance decrease slowly and EXT4 performance increase. Figure2 Shows with large file size JFS has higher performance.

Figure3: IOzone test for Write feature Figure 1: IOzone test for Read feature According to above explanation, Figure1 shows that EXT4 has weakness in small file. Also this graph shows that ReiserFS and EXT3 have a good performance with small file. But with increasing file size, EXT4 performance increase as well. Also this graph shows that JFS more stable than other file system.

Figure3 shows write performance in these four file system. Figure3 shows that EXT4 and JFS has a good performance with small file size. ReiserFS has the lowest performance. But with increasing file size the EXT4 performance is not change too much, but JFS performance decreasing too much. In large file size JSF has the worst performance.

Figure 4: IOzone test for Re-Write feature Figure4 shows Re-write feature that tested by four file system. For this feature, JFS has the best performance base on small file size and large file size. This graph shows that EXT4 performance slowly increase and JFS performance deeply decrease.

Figure 2: IOzone test for Re-Read feature

Figure5: All feature performance for small file size (1MB). Figure7: Postmark test result base on creation, append, deletion, and read.

1. CONCLUSION AND FUTURE WORK

Figure6: All feature performance for large file size (100MB). Figure 5 and Figure 6 shows other perspective of performance of these four file system. With comparison four feature performance in small file size; JFS has a good performance in Re-Write, and Re-Read. Also the worth performance in small file size belong to EXT4. Comparison between EXT3 and EXT4 which are using extent, EXT3 is much better than EXT4. Also according to large file size, EXT4 mostly has a better performance than others. Specially, on their write file. JFS also has a good performance on read file. Postmark result According to postmark description, run postmark on four file systems. (EXT4, JFS, EXT3, and ReiserFS) Postmark used maximum create, read, append, and delete files base on Table 1. Table1: Four file systems comparison table with postmark.

Choose a file system to managing data on your system, is one of the consideration for anybody who use a computer. Anybody can fill this problem when you have a lot of files with a large portion [20]. This project define base on famous file system comparison on Linux. Project starts which a research on file system on Linux and then find some research on these file systems. There is some evaluator tools look like FFSB, IOzone, and Postmark. Benchmark which using IOZONE and POSTMORK which is more famous. These tools are choosing to check which one is more reliable for evaluation for my project. Some of the Research objectives in this project are carrying out as follow (i) Assess different type of FS on Linux Assess different type of file system performance evaluator tools (ii) Categorizes file system base in performance. The result shows that EXT4 which is using extend have a performance on large file size but it is not suitable for small file size. In other word, with increase file size, the EXT4 performance increase. Other file systems (which is JFS and using journaling) also has a good performance (especially in write and Re-write file). But JFS is not update any more during 2 years. EXT3 also has a reliable performance on small file size.

9

Reference

[1] Bryant, R., Forester, R., & Hawkes, J. (n.d.). File system Performance and Scalability in Linux 2.4.17. Proceedings of the FREENIX Track:2002 USENIX Annual Technical Conference. USENIX. [2] Sun Microsystems (2004). File System Performance: The Solaris™ OS, UFS, Linux EXT3, and ReiserFS. Aug.

Figure7 shows result of postmark testing. This graph shows that EXT4 is the fastest file system in all of the execution part. Also ReiserFS are in the second level.

[3] How to Find the Block Size. (2005, Aug 18). Retrieved Oct 17, 2009, from LINFO: http://www.linfo.org/get_block_size.html

[4] Terminal Window Definition. (2005, May 1). Retrieved Nov 9, 2009, from The Linux Information Project: http://www.linfo.org/terminal_window.html

[19] Norcott., W. (98). Iozone Filesystem Benchmark. Retrieved Sep 15, 2009, from iozone: http://www.iozone.org/docs/IOzone_msword_98.pdf

[5] Inode Definition. (2006, Sep 15). Retrieved Jan 5, 2010, from The Linux Information Project: http://www.linfo.org/inode.html

[20] Stepohen, S. (2010). Novell makes file storage software shift. Retrieved Dec 23, 2009, from

[6] Boyne, J. (2005). Disc and Volume Size Limits. [7] Diedrich, O. (2009, May 29). The Ext4 Linux File System. Retrieved Dec20,2009,from the Open:http://www.honline.com/open/features/The-Ext4-Linux-file-system746579.html [8] Henson, V., Brown, Z., Ts’o, T., & van de Ven, A. (2006). Reducing fsck time for EXT2 file systems. Linux Symposium, p. 395. [9] Ts'o, T. (2002). Planned extensions to the Linux EXT2/EXT3 File system. USENIX 2002 Annual Technical Conference, Freenix Track , pp. 235–244 . [10] Tweedie, S. (98). Journaling the Linux EXT2fs File system. LinuxExpo. [11] Y. Ts'o , T., & Stephen, T. (2002, June 10). Planned extensions to the Linux EXT2/EXT3 File system . USENIX Association. [12] Layton, J. (2009, March 28). EXT4 File System: Introduction and Benchmarks. Retrieved Dec 29, 2009, from Linux mag: http://www.linux-mag.com/id/7271/1 [13] Norcott., W. (98). IOzone File system Benchmark. Retrieved Sep 15, 2009,from IOzone: ttp://www.IOzone.org/docs/IOzone_msword_98.pdf [14] Katcher, J. (97, 10 8). Postmark: A New File System Benchmark. Retrieved Dec 1, 2009, from http://www.netapp.com/technology/level3/3022.html: http://communities.netapp.com/servlet/JiveServlet/download/ 2609-1551/Katcher97-postmark-netapp[15] Y. Ts'o , T., & Stephen, T. (2002, June 10). Planned Extensions to the Linux Ext2/Ext3 Filesystem . USENIX Association. [16] Tweedie, S. (98). Filesystem. LinuxExpo.

Journaling

the

Linux

ext2fs

[17] Galli Granada, P. (2002, Jan 1). Journal File Systems in Linux. p. 6. [18] Steve. (2000, Jan 1). JFS overview . Best works in the Software Solutions & Strategy Division of IBM in Austin.